AnandTech reader Grant Vezina pointed out in the comments to my last post that Intel properly documents almost all Ivy Bridge die sizes in the mechanical specifications pages of its Ivy Bridge datasheets. That's what I get for feeling accomplished after taking apart the Zenbook Prime and measuring its Core i7-3517U. Only the 4C/GT2, 2C/GT2 and 2C/GT1 dimensions are documented while the 4C/GT1 configuration is Vezina's own calculation based on available data. 

Ivy Bridge Die Comparison
CPU CPU Cores L3 Cache GPU Config Dimensions Die Area
Ivy Bridge HE-4 4 8MB GT2 8.141 x 19.361 mm 159.8mm2
Ivy Bridge HM-4 4 6MB GT1 7.656 x 17.349 mm 132.8mm2
Ivy Bridge H-2 2 4MB GT2 8.141 x 14.505 mm 118.1mm2
Ivy Bridge M-2 2 3MB GT1 7.656 x 12.223 mm 93.6mm2

There are a few items of interest here. The GT1 configs appear to be narrower than the GT2 counterparts, otherwise these chips just differ in terms of die length. Haswell will likely continue the trend. 

Why did Intel stop at 16 EUs for the GT2 in Ivy Bridge? Cost is an obvious concern (Intel likes making tons of money) but at some point you need to scale up memory bandwidth to make use of additional compute horsepower. Haswell will address this issue while it scales up to 40 EUs. Intel could have implemented eDRAM with Ivy Bridge, however only one customer was really asking for it and wasn't interested in paying a significant premium for it. Why pay now when Haswell will deliver it less than a year later?

POST A COMMENT

14 Comments

View All Comments

  • JKflipflop98 - Saturday, June 02, 2012 - link

    A lot of other places use defective parts with sections disabled as the lower-tiered parts. Intel doesn't mess up enough die to keep up with demand for the lower-price parts (which always do way more volume than the high-end parts). So instead of wasting die space on a part that you know you're going to purposely gimp in order to sell for less, we just make the smaller die with the low-end specifications.

    You get the part you wanted at the price you wanted, and Intel gets to put extra die on a wafer. It's a win-win.
    Reply
  • stadisticado - Saturday, June 02, 2012 - link

    Oh there surely is still binning going on, we just haven't seen those products yet. Just think, we have four total die for desktop/mobile. I'm sure once the Core i3/Pentium/Celeron IVB variants get released you'll see that those are definitely downbinned versions of these four with defective EU units/cache/low speed etc. Reply
  • ajp_anton - Monday, June 04, 2012 - link

    Intel rarely harvests dice. I talked to one of their representatives at a computer event once, and he said they just throw away non-functional chips as it's not worth the effort. If they disable parts, it's only for artificial market segmentation and not because of harvesting.
    I don't know how much he's exaggerating, but I wouldn't be surprised considering how good their manufacturing is.
    Reply
  • CeriseCogburn - Monday, June 04, 2012 - link

    Whereas AMD has a long running list of identical cores with disabled aka crippled parts belched out as standard operating procedure.

    I wonder if there have been more than 3 AMD chips with a hundred different cpu's sometimes.

    Talk about rebranding, AMD cpus are the absolute king of rebranding, for years running.
    Reply

Log in

Don't have an account? Sign up now